Explainable Alzheimer's Disease Detection Using Linguistic Features from Automatic Speech Recognition.

Area of Science:

• Neurology
• Computational Linguistics
• Artificial Intelligence

Background:

• Alzheimer's disease (AD) is a leading cause of dementia, characterized by cognitive decline and loss of life skills.
• Early detection of AD is crucial for effective intervention and management.
• Speech dysfunction is an early indicator of AD, making speech analysis a potential screening tool.

Purpose of the Study:

• To investigate the effectiveness of automatic speech recognition (ASR) for building an end-to-end automated speech analysis model for AD detection.
• To evaluate ASR's efficiency compared to manual transcription in extracting linguistic features for AD assessment.

Main Methods:

• Three publicly available ASR engines were implemented and evaluated on the ADReSS-IS2020 dataset.
• Classification performance was compared between automated and manual transcriptions.
• The SHapley Additive exPlanations algorithm was used to identify key linguistic features influencing model performance.

Main Results:

• ASR tools achieved mean word error rates between 32% and 43%.
• Automated transcriptions yielded classification accuracies ranging from 81.25% to 89.58%, comparable or superior to manual transcriptions.
• The best model, utilizing ensemble learning, demonstrated performance on par with state-of-the-art manual transcription-based methods.

Conclusions:

• An end-to-end automated speech analysis system for AD detection using ASR is feasible.
• ASR offers an efficient alternative to manual transcription for AD screening.
• Identified critical linguistic features may offer insights into AD mechanisms.